Randomized trial compares the Marquardt algorithm's efficiency in function approximation tasks using feedforward neural networks, highlighting superior performance.
Key Points
This research aims to evaluate the efficiency of the Marquardt algorithm for training feedforward neural networks compared to other techniques.
Incorporated the Marquardt algorithm into backpropagation for training neural networks.
Tested the algorithm on several function approximation problems.
Compared performance with conjugate gradient and variable learning rate algorithms.
The Marquardt algorithm showed much greater efficiency than both conjugate gradient and variable learning rate algorithms.
Performance advantages were particularly notable for networks with a few hundred weights.